Gartner: AI PCs could finally matter by slashing token costs

Gartner’s pitch: AI PCs as a hedge on token bills

Corporate buyers have mostly shrugged at AI PCs, but Gartner now argues they may have a very specific use: cutting runaway cloud token costs.

In a new Strategic Roadmap for Agentic AI PCs, Gartner Research VP Steve Kleynhans positions AI-capable desktops and laptops as a key part of a hybrid AI strategy, offloading the right workloads from the cloud to the client.

“On-device AI has yet to achieve mainstream adoption and remains largely confined to developer and enthusiast use cases,” he wrote, adding that desktop-focused enterprise tools “have been slow to materialize.”

Rising tokenomics pain is the driver

Kleynhans says interest is shifting as enterprises get a clearer view of cloud AI economics, especially so‑called “Tokenomics” — the messy variability in how providers define and bill for tokens.

Recommended reading

Mira Murati’s lab unveils Inkling, a 975B open model

“As enterprises gain a better understanding of AI cloud cost dynamics, many are looking to AI PCs as a potential offset,” he wrote.

According to Gartner, organizations are increasingly worried about the sustainability of cloud‑only AI as token consumption and usage fees climb. That’s pushing them toward a hybrid strategy that identifies workloads better handled at the edge or directly on the device.

Kleynhans concedes “there is no consensus yet on the level of cost benefit” but maintains the savings potential is “clear.”

Small models on today’s hardware

The optimism rests on the rise of:

• Small language models (SLMs)
• Small reasoning models (SRMs)
• Targeted domain-specific language models

Kleynhans expects some of these to run comfortably on current AI PCs, which ship with neural processing units (NPUs) capable of at least 50 TOPS.

He’s not alone; as The Register has reported, Microsoft and Google are also leaning on smaller models for certain jobs.

Enterprise-ready tools are still immature, but Kleynhans points to projects such as OpenClaw and on-device offerings like Claude Cowork, Microsoft Scout, and OpenAI Codex as early proof points that can show buyers what’s feasible on local hardware.

From endpoint to AI node

Kleynhans forecasts that local AI models will underpin on-device speech, chat, image, audio, and text generation, plus application and model orchestration.

“SLMs and SRMs will power always-on personal assistants and agents, fundamentally changing how users interact with their devices.”

He expects many routine tasks to run locally, with personal agents orchestrating work across apps, models, and services that span both device and cloud.

Cloud infrastructure doesn’t disappear in this vision. It still hosts the heaviest workloads, while mature models gradually move down to endpoints as they’re optimized, “transforming the PC from a simple endpoint into a critical component of the broader AI infrastructure.”

Gartner also sees a steady hardware ramp, with AI PCs becoming ten times more powerful by 2031.

Gartner’s timelines and advice

Kleynhans offers two concrete predictions:

• By 2029, 30 percent of enterprises will use AI PCs to reduce cloud AI token costs.
• By 2030, 70 percent of the corporate PC installed base will be able to run some local GenAI workloads.

To prepare, he urges organizations to treat AI PCs as part of the core IT infrastructure, not just a nicer client device. That means building an ROI model explicitly tied to displacing cloud token spend.

“Initially this will be most appropriate for developers but make the discussion part of any new AI deployment for all employees,” he suggests.

Kleynhans recommends ramping this effort in earnest with third‑gen AI PCs, which he expects to arrive in 2027, and starting experiments now with SLMs and SRMs on the hardware already available.